What are Tokens? A Simple Introduction to AI Pricing and 'Limit Reached' Errors
When you start using AI, many people have these questions at least once.
"How is the pricing calculated?"
"Why does it get slower when I input long text?"
"What exactly is the 'limit' in 'Limit Reached'?"
These are not separate issues.
The answer to all of them lies in a common AI unit called a "token."
By reading this article, you will understand the following three things:
What a token is — The story of the "chunks" AI uses to read text
How AI pricing is determined — Why both input and output cost money
Why "Limit Reached" happens — And how to avoid it
This article is for non-engineers who use AI tools for work. It is not a technical explanation for developers, but aims to help you "understand your billing" and "prevent interruptions."
This is a follow-up to the previous "What is an API?" article, but it can be read on its own.
Conclusion: Tokens are the "character count" of the AI world
A token is the smallest chunk AI uses when processing text. It is a unit similar to character count.
AI pricing is pay-as-you-go based on tokens. Furthermore, it applies to both input (what you send) and output (what is returned).
There is a limit to the number of tokens that can be handled at once, and exceeding it causes "Limit Reached" errors or memory loss.
What is a token?
AI does not read text as it is; it breaks it down into small chunks before processing. These chunks are tokens.
The important thing is that Japanese uses more tokens than English. Even if you convey the same content, Japanese tends to be more expensive. For those of us using it in Japanese, this is a fact that quietly adds up.
Pricing applies to both "input" and "output"
This is the most important part to understand when reading your bill.
In fact, every company's pricing page lists costs separately as "input $X / output $X" (OpenAI / Google Gemini / Anthropic, all as of July 28, 2026).
As you can see, the unit price for output is generally higher. In other words, the more you make it write, the more it costs. Asking it to write a long text is more expensive than asking it to summarize.
See for yourself
It is faster to try it out yourself. If you paste text into the Tokenizer page published by OpenAI, it will count how many tokens it is on the spot. It is free and does not require a login.
Try pasting one of the emails you usually write. Once you see "how many tokens your own writing is," the numbers on the pricing page will suddenly feel real.
The truth behind "limit reached"
Another common frustration can also be explained by tokens.
AI has a limit on the amount of information it can spread out on its desk at once (context). As you continue a conversation, past exchanges pile up on the desk. When the desk is full, it will either stop or start forgetting the oldest parts of the conversation.
Longer conversations do not necessarily make the AI smarter. In fact, it is the opposite. They become slower, more expensive, and less accurate. If the topic changes, open a new chat. That alone will improve things significantly.
How this applies to work
Here are two specific examples.
One is meeting minutes summarization. If you paste the entire transcript of a 60-minute meeting, the input tokens will balloon at once. By simply changing your approach to "paste only the decision-making parts," you can improve both cost and accuracy. AI is more accurate when you narrow down the information rather than having it read everything.
The other is automated inquiry classification. Even if each one is short, if you process thousands per month, the total adds up. For this kind of "short but high-volume" processing, you rarely need to use a top-tier model. Letting a lighter model handle it can reduce costs to a fraction.
In a Forvio verification article, when we tested the same classification, translation, and summarization tasks with a lighter model, we were able to significantly reduce costs without sacrificing quality ("Can Gemini 3.5 Flash-Lite be used for high-volume processing?"). Once you understand tokens, you will be able to make decisions about "which model to use for what task."
Savings you can start today
No settings changes are required for any of these. They work just by changing how you use the tools.
You can check how much you are actually using in the management console of each service (e.g., Google AI Studio usage dashboard).
To avoid surprises at the end of the month, it is safer to set a limit in advance.
Frequently Asked Questions
What are tokens in simple terms? They are chunks of text that AI uses to process information. In Japanese, one character is generally equivalent to a little over one token, so it is safe to think of them as a unit close to the character count.
How is AI pricing calculated? It is (number of input tokens × input unit price) + (number of output tokens × output unit price). Unit prices are generally displayed per 1 million tokens and can be checked on each company's official pricing page.
Why do I get a 'token limit reached' message? This is because you have exceeded the amount of tokens that can be handled in a single conversation. Please start a new chat or narrow down the material you are pasting to only the necessary parts.
Is it true that Japanese is more expensive than English? For the same content, Japanese tends to use more tokens. As a result, the cost is often higher.
Do tokens matter for monthly chat plans? Yes, they do. Even with monthly plans, token limits are active internally, which can cause long conversations to stop. While it is different from a 'pay-as-you-go' model, the limit itself still exists.
Does quality drop if I reduce tokens? Not necessarily. In fact, AI answers are often more accurate when you remove unnecessary information and provide only the essential parts.
Summary
Tokens are the character count of the AI world.
Pricing, limits, and speed are all determined by this unit.
While the previous article on 'APIs' was about 'connecting AI to the outside world,' this article on tokens was about the 'cost of running AI.' Once you understand these two, you can read almost any AI tool's pricing page and settings screen.
Next time, we will cover what to ask the AI and how—the **'prompt.'**
For those who want to learn more systematically
If you want to grasp the basic IT/AI terms like API, cloud, SaaS, and generative AI all at once, an introductory book on the IT Passport exam is the shortcut. It organizes the terms you encounter at work in one volume for non-engineers (items related to generative AI have also been added to the IT Passport exam syllabus / IPA).
IT Passport introductory books
*This article contains promotions (affiliate links).
Related Articles
What is an API? An easy-to-understand explanation (AI Basics Series Part 1)
Can Gemini 3.5 Flash-Lite be used for high-volume processing? Verifying translation, summarization, and classification in 3 cases
[August 2026 Edition] Comparing Gemini, Claude, and ChatGPT on the same tasks
Thank you for reading to the end.
If you found this article helpful, your support through likes and follows would be greatly appreciated.
Please also check out Forvio, where we test and share how to utilize AI in practical work.